VNIR Spectroscopy Estimates of Within-Field Variability in Soil Properties

Abstract

Over the last three decades or more, researchers have estimated soil properties using visible and near-infrared (VNIR) diffuse reflectance spectroscopy (DRS), with varying results. Using VNIR DRS for estimating soil property variation within fields is particularly challenging, because in many cases the variation in the property of interest may be relatively small and because of the need to deal with spatially correlated data. In this study, we used VNIR DRS to estimate the variability in soil physical and chemical properties within a typical production field in north-east Missouri, USA. Soil samples were obtained to 15 cm depth on a 30 m grid spacing, plus at a number of random sampling locations. Laboratory analyses were conducted for sand, silt, and clay fractions, organic matter, pH, P, K, Ca, Mg, and cation exchange capacity (CEC). VNIR reflectance of dried and sieved samples was obtained in the laboratory using a spectrometer with a wavelength range from 350 to 2,500 nm. Partial least squares (PLS) regression was used to estimate soil properties from spectra, both for the full range and for subset wavelength ranges above and below 1,000 nm. A regression kriging approach was used to account for spatial dependence. We found that for these soils, an NIR-only instrument (1,000–2,500 nm) would be able to quantify CEC, organic matter, and texture with accuracy similar to that from a VNIR (350–2,500 nm) instrument. Some soil properties, including CEC and pH, were well estimated with PLS regression, while others, including organic matter, were not. Coupling regression kriging with PLS regression improved estimates in some cases, but was not as effective as has been reported in some other studies. More advanced calibration methods should be investigated for their ability to improve within-field VNIR DRS results on these soils.